Replaces default model priors with user specified ones
(restricted to normal priors with specified mean and standard
deviations).
A common use is extracting the posterior from a previous
epinowcast() run (using summary(nowcast, type = "fit"))
and using it as a prior for subsequent fits.
Arguments
- priors
A
data.framewith columnsvariable,mean, andsddescribing normal priors. Default priors in the appropriate format are returned by the$priorselement ofenw_reference()and other model module functions.- custom_priors
A
data.framewith columnsvariable,mean, andsddescribing normal priors to use as replacements. Rows incustom_priorsreplace matching rows inpriorsby thevariablecolumn. Vectorised prior names (i.e. those of the formvariable[n]) are matched after stripping the index (treated asvariable).
Details
Default priors can be obtained from each model module's
$priors element, e.g. enw_reference(data = pobs)$priors.
See the priors argument of epinowcast() for a list of
available prior variable names by module.
See also
Functions used to help convert models into the format required for stan
enw_formula_as_data_list(),
enw_get_cache(),
enw_model(),
enw_pathfinder(),
enw_priors_as_data_list(),
enw_sample(),
enw_set_cache(),
enw_stan_to_r(),
enw_unset_cache(),
remove_profiling(),
write_stan_files_no_profile()
Examples
# Update priors from a data.frame
priors <- data.frame(variable = c("x", "y"), mean = c(1, 2), sd = c(1, 2))
custom_priors <- data.frame(variable = "x[1]", mean = 10, sd = 2)
enw_replace_priors(priors, custom_priors)
#> variable mean sd
#> <char> <num> <num>
#> 1: y 2 2
#> 2: x 10 2
# Update priors from a previous model fit
default_priors <- enw_reference(
distribution = "lognormal",
data = enw_example("preprocessed"),
)$priors
print(default_priors)
#> variable
#> <char>
#> 1: refp_mean_int
#> 2: refp_sd_int
#> 3: refp_mean_beta_sd
#> 4: refp_sd_beta_sd
#> 5: refp_arima_sigma
#> 6: refp_arima_sd_sigma
#> 7: refp_arima_pacf
#> 8: refp_gp_rho
#> 9: refp_gp_alpha
#> 10: refp_gp_sd_alpha
#> 11: refnp_int
#> 12: refnp_beta_sd
#> 13: refnp_arima_sigma
#> 14: refnp_arima_pacf
#> 15: refnp_gp_rho
#> 16: refnp_gp_alpha
#> description
#> <char>
#> 1: Log mean intercept for parametric reference date delay
#> 2: Log standard deviation for the parametric reference date delay
#> 3: Standard deviation of scaled pooled parametric mean effects
#> 4: Standard deviation of scaled pooled parametric sd effects
#> 5: Scale of the ARIMA latent residual on the parametric reference mean
#> 6: Scale of the ARIMA latent residual on the parametric reference sd
#> 7: Partial autocorrelations of the ARIMA latent residual on the parametric reference; Uniform(-1, 1) when sd = 0, otherwise Normal(mean, sd) truncated to (-1, 1)
#> 8: Length scale of the Gaussian process on the parametric reference; log-normal prior on the (positive) length scale
#> 9: Magnitude (marginal standard deviation) of the Gaussian process on the parametric reference mean; half-normal prior
#> 10: Magnitude (marginal standard deviation) of the Gaussian process on the parametric reference sd; half-normal prior
#> 11: Intercept for non-parametric reference date delay
#> 12: Standard deviation of scaled pooled non-parametric effects
#> 13: Standard deviation of the ARIMA latent residual on non-parametric reference logit hazards
#> 14: Partial autocorrelations of the ARIMA latent residual on the non-parametric reference; Uniform(-1, 1) when sd = 0, otherwise Normal(mean, sd) truncated to (-1, 1)
#> 15: Length scale of the Gaussian process on the non-parametric reference; log-normal prior on the (positive) length scale
#> 16: Magnitude (marginal standard deviation) of the Gaussian process on the non-parametric reference; half-normal prior
#> distribution mean sd
#> <char> <num> <num>
#> 1: Normal 1.000000 1.00
#> 2: Zero truncated normal 0.500000 1.00
#> 3: Zero truncated normal 0.000000 1.00
#> 4: Zero truncated normal 0.000000 1.00
#> 5: Zero truncated normal 0.000000 0.20
#> 6: Zero truncated normal 0.000000 0.20
#> 7: Uniform 0.000000 0.00
#> 8: Log normal 1.098612 0.50
#> 9: Zero truncated normal 0.000000 0.05
#> 10: Zero truncated normal 0.000000 0.05
#> 11: Normal 0.000000 1.00
#> 12: Zero truncated normal 0.000000 1.00
#> 13: Zero truncated normal 0.000000 0.20
#> 14: Uniform 0.000000 0.00
#> 15: Log normal 1.098612 0.50
#> 16: Zero truncated normal 0.000000 0.05
fit_priors <- summary(
enw_example("nowcast"),
type = "fit",
variables = c("refp_mean_int", "refp_sd_int", "sqrt_phi")
)
fit_priors
#> variable mean median sd mad q5 q20
#> <char> <num> <num> <num> <num> <num> <num>
#> 1: refp_mean_int[1] 2.2845186 2.265566 0.49001224 0.47245807 1.522838 1.868480
#> 2: refp_sd_int[1] 3.2060377 3.201288 0.22387183 0.23731815 2.853391 3.008119
#> 3: sqrt_phi[1] 0.2496831 0.248297 0.03409312 0.03389633 0.196707 0.221606
#> q80 q95 rhat ess_bulk ess_tail
#> <num> <num> <num> <num> <num>
#> 1: 2.6828413 3.1463897 1.006471 571.2456 635.6078
#> 2: 3.3975243 3.5802367 1.005243 686.0466 746.0435
#> 3: 0.2784192 0.3063322 1.001405 1086.0941 603.1580
enw_replace_priors(default_priors, fit_priors)
#> variable
#> <char>
#> 1: refp_mean_beta_sd
#> 2: refp_sd_beta_sd
#> 3: refp_arima_sigma
#> 4: refp_arima_sd_sigma
#> 5: refp_arima_pacf
#> 6: refp_gp_rho
#> 7: refp_gp_alpha
#> 8: refp_gp_sd_alpha
#> 9: refnp_int
#> 10: refnp_beta_sd
#> 11: refnp_arima_sigma
#> 12: refnp_arima_pacf
#> 13: refnp_gp_rho
#> 14: refnp_gp_alpha
#> 15: refp_mean_int
#> 16: refp_sd_int
#> 17: sqrt_phi
#> description
#> <char>
#> 1: Standard deviation of scaled pooled parametric mean effects
#> 2: Standard deviation of scaled pooled parametric sd effects
#> 3: Scale of the ARIMA latent residual on the parametric reference mean
#> 4: Scale of the ARIMA latent residual on the parametric reference sd
#> 5: Partial autocorrelations of the ARIMA latent residual on the parametric reference; Uniform(-1, 1) when sd = 0, otherwise Normal(mean, sd) truncated to (-1, 1)
#> 6: Length scale of the Gaussian process on the parametric reference; log-normal prior on the (positive) length scale
#> 7: Magnitude (marginal standard deviation) of the Gaussian process on the parametric reference mean; half-normal prior
#> 8: Magnitude (marginal standard deviation) of the Gaussian process on the parametric reference sd; half-normal prior
#> 9: Intercept for non-parametric reference date delay
#> 10: Standard deviation of scaled pooled non-parametric effects
#> 11: Standard deviation of the ARIMA latent residual on non-parametric reference logit hazards
#> 12: Partial autocorrelations of the ARIMA latent residual on the non-parametric reference; Uniform(-1, 1) when sd = 0, otherwise Normal(mean, sd) truncated to (-1, 1)
#> 13: Length scale of the Gaussian process on the non-parametric reference; log-normal prior on the (positive) length scale
#> 14: Magnitude (marginal standard deviation) of the Gaussian process on the non-parametric reference; half-normal prior
#> 15: <NA>
#> 16: <NA>
#> 17: <NA>
#> distribution mean sd
#> <char> <num> <num>
#> 1: Zero truncated normal 0.0000000 1.00000000
#> 2: Zero truncated normal 0.0000000 1.00000000
#> 3: Zero truncated normal 0.0000000 0.20000000
#> 4: Zero truncated normal 0.0000000 0.20000000
#> 5: Uniform 0.0000000 0.00000000
#> 6: Log normal 1.0986123 0.50000000
#> 7: Zero truncated normal 0.0000000 0.05000000
#> 8: Zero truncated normal 0.0000000 0.05000000
#> 9: Normal 0.0000000 1.00000000
#> 10: Zero truncated normal 0.0000000 1.00000000
#> 11: Zero truncated normal 0.0000000 0.20000000
#> 12: Uniform 0.0000000 0.00000000
#> 13: Log normal 1.0986123 0.50000000
#> 14: Zero truncated normal 0.0000000 0.05000000
#> 15: <NA> 2.2845186 0.49001224
#> 16: <NA> 3.2060377 0.22387183
#> 17: <NA> 0.2496831 0.03409312